Commits

Branch selector

 master

User selector

All users

Datepicker

All time

Commit History

Commits on Nov 1, 2025

• sync : llama.cpp

  ggerganov

  committed
  09aa758

  Copy full SHA for 09aa758

• CUDA: add expert reduce kernel (llama/16857)

  Show description for 82b8e26
  

  2 peopleauthored and

  ggerganov

  committed
  82b8e26

  Copy full SHA for 82b8e26

• vulkan: disable spirv-opt for rope shaders (llama/16872)

  

  jeffbolznv

  authored and

  ggerganov

  committed
  a4ec1c5

  Copy full SHA for a4ec1c5

• vulkan: Fix crash when FP16 mul_mat accumulation is not supported (llama/16796)

  Show description for bef4e1d
  

  rillomas

  authored and

  ggerganov

  committed
  bef4e1d

  Copy full SHA for bef4e1d

• vulkan: fix shmem overrun in mmq id shader (llama/16873)

  Show description for d029dac
  

  0cc4m

  and

  ggerganov

  committed
  d029dac

  Copy full SHA for d029dac

• ggml-hexagon: respect input size when getting/setting tensor data (llama/16836)

  Show description for 6adf9f7
  

  l3utterfly

  authored and

  ggerganov

  committed
  6adf9f7

  Copy full SHA for 6adf9f7

• opencl: fix boundary handling for mul_mm (llama/16875)

  

  lhez

  authored and

  ggerganov

  committed
  f14754d

  Copy full SHA for f14754d

• cpu: introduce chunking for repack matmuls and enable matmul-id chunking on ARM64 (llama/16833)

  Show description for bf54537
  

  max-krasnyansky

  authored and

  ggerganov

  committed
  bf54537

  Copy full SHA for bf54537

• model: add support for qwen3vl series (llama/16780)

  Show description for fd33b4b
  

  7 peoplecommitted
  fd33b4b

  Copy full SHA for fd33b4b

• cpu: introduce chunking for flash attention (llama/16829)

  Show description for d508a2b
  

  max-krasnyansky

  authored and

  ggerganov

  committed
  d508a2b

  Copy full SHA for d508a2b

• cuda : fix argsort with 64k+ rows (llama/16849)

  

  CISC

  authored and

  ggerganov

  committed
  5f7ee94

  Copy full SHA for 5f7ee94

• vulkan: Handle argsort with a large number of rows (llama/16851)

  

  jeffbolznv

  authored and

  ggerganov

  committed
  017ad2f

  Copy full SHA for 017ad2f

• Hide latency of bias and gate-loading (llama/16847)

  Show description for a66c591
  

  ORippler

  authored and

  ggerganov

  committed
  a66c591

  Copy full SHA for a66c591

• vulkan: Fuse rope+set_rows (llama/16769)

  Show description for 0f0fd00
  

  jeffbolznv

  authored and

  ggerganov

  committed
  0f0fd00

  Copy full SHA for 0f0fd00

• vulkan: Update topk_moe fusion to handle gpt's late softmax (llama/16656)

  Show description for 54ab43b
  

  2 peopleauthored and

  ggerganov

  committed
  54ab43b

  Copy full SHA for 54ab43b

• Vulkan MMQ Integer Dot Refactor and K-Quant support (llama/16536)

  Show description for 558148d
  

  0cc4m

  authored and

  ggerganov

  committed
  558148d

  Copy full SHA for 558148d

• Hexagon Op queue & dispatch optimizations (llama/16820)

  Show description for 4f45595
  

  max-krasnyansky

  authored and

  ggerganov

  committed
  4f45595

  Copy full SHA for 4f45595

• CUDA: use fastdiv in set-rows (llama/16834)

  Show description for 9f3fde1
  

  am17an

  authored and

  ggerganov

  committed
  9f3fde1

  Copy full SHA for 9f3fde1

• vulkan: Call ggml_vk_buffer_write_2d from ggml_vk_buffer_copy (llama/16793)

  Show description for 33a1faf
  

  jeffbolznv

  authored and

  ggerganov

  committed
  33a1faf

  Copy full SHA for 33a1faf

• CUDA: Fix bug in topk-moe for gpt-oss (llama/16821)

  Show description for 52589d7
  

  am17an

  authored and

  ggerganov

  committed
  52589d7

  Copy full SHA for 52589d7

• sycl: add RMS_NORM_BACK operation support (llama/16808)

  Show description for 2141745
  

  2 peopleauthored and

  ggerganov

  committed
  2141745

  Copy full SHA for 2141745

• cuda: add SET operation support (llama/16804)

  Show description for 7a68370
  

  2 peopleauthored and

  ggerganov

  committed
  7a68370

  Copy full SHA for 7a68370

• initialise buffer.device in ggml_hexagon_session (llama/16816)

  

  l3utterfly

  authored and

  ggerganov

  committed
  cfa65d6

  Copy full SHA for cfa65d6

• CANN: Improve device ID handling and aclnnArange checks (llama/16752)

  Show description for 927d8c5
  

  noemotiovon

  authored and

  ggerganov

  committed
  927d8c5

  Copy full SHA for 927d8c5

• CUDA: add unused vars to mmvf and mmvq (llama/16807)

  

  am17an

  authored and

  ggerganov

  committed
  1310026

  Copy full SHA for 1310026

• sycl: add SSM_CONV operation support (llama/16800)

  Show description for 097c6b4
  

  2 peopleauthored and

  ggerganov

  committed
  097c6b4

  Copy full SHA for 097c6b4

• ggml : fix interpolate with align-corners and ne=1 (llama/16700)

  Show description for da6b57b
  

  Acly

  authored and

  ggerganov

  committed
  da6b57b

  Copy full SHA for da6b57b

• HIP: fix AMDGPU_TARGETS, update documentation (llama/16803)

  

  JohannesGaessler

  authored and

  ggerganov

  committed
  006482f

  Copy full SHA for 006482f

• test-backend-ops: print failed tests at the end (llama/16785)

  

  am17an

  authored and

  ggerganov

  committed
  90ab6b2

  Copy full SHA for 90ab6b2

• sycl: add ROLL operation support (llama/16665)

  Show description for 8a74d55
  

  2 peopleauthored and

  ggerganov

  committed
  8a74d55

  Copy full SHA for 8a74d55

• sycl: add REPEAT_BACK operation support (llama/16734)

  Show description for a25b01e
  

  2 peopleauthored and

  ggerganov

  committed
  a25b01e

  Copy full SHA for a25b01e

• CUDA: support for weight clamp in top-k norm (llama/16702)

  

  am17an

  authored and

  ggerganov

  committed
  720d0fb

  Copy full SHA for 720d0fb

• ggml-alloc : make gallocr prefer chunks that allow memory reuse (llama/16788)

  

  Acly

  authored and

  ggerganov

  committed
  058128a

  Copy full SHA for 058128a

• cuda : use fast copy when src and dst are of different type and contiguous (llama/16789)

  Show description for b3f54b8
  

  CISC

  authored and

  ggerganov

  committed
  b3f54b8

  Copy full SHA for b3f54b8

• ggml: fix cuda kernel launch configuration for k_compute_batched_ptrs to support large batch (llama/16744)

  Show description for ebc4993
  

  2 peopleauthored and

  ggerganov

  committed
  ebc4993

  Copy full SHA for ebc4993

Pagination

Previous

Next